fix(kio): split waiters by condition so writes don't churn closed/consumer waiters

[kixelated]

Summary

kio used a single waiter list, so every event woke every parked waiter. The expensive case is a per-modification write (the hot path): it woke closed(), used(), and unused() waiters that don't care about value changes, and conversely consumer create/drop woke the value waiters. Those spurious wakes just re-register and ping-pong, which is the root cause behind the per-poll consume() churn.

This splits State into three lists, keyed by what each waiter actually waits on:

List	Waiters	Woken by
waiters_value	poll/wait	every modification (hot path) + close
waiters_closed	closed()	close only
waiters_consumer	used()/unused()	consumer 0↔1 + close

Close still wakes all three, since every waiter resolves on close. No other event over-wakes.

Why these groupings

• used/unused share one list. Every call site uses them sequentially (used().await, then a loop selecting on unused()), so they're never parked on the same channel simultaneously and a combined list never cross-wakes in practice. A separate unused list would add cost for no benefit.
• closed() gets its own list. It's heavily used (87 call sites) and parked long-term in select! blocks. Sharing the value list meant it ate a spurious wake on every frame write — a bigger churn source than used/unused.

Also

• Added a [Unreleased] CHANGELOG entry.
• Added waker-counting regression tests asserting each event wakes only the waiters that care: value writes don't wake closed()/unused(), consumer churn doesn't wake value/closed() waiters, and close wakes all three.

Relation to #1735

Rebased on top of #1735 (read-only-predicate poll/wait rework). That PR fixed the "a no-op mutation during a pending poll wakes the poller's own waiter" footgun by making the predicate read-only; this PR fixes the orthogonal "every event wakes every waiter" problem by splitting the list. The two compose: poll/wait now register in waiters_value, and a write through the returned Mut wakes only that list (plus closed/consumer on close).

Public API

No public API change — State and its waiter lists are pub(crate). Internal behavior-preserving fix, hence targeting main.

Tradeoff

State now holds three WaiterLists (each ~256 B inline via INLINE_WAITERS = 32), so it grows ~512 B vs. the single-list original. waiters_value/waiters_closed both scale with subscriber count, but waiters_consumer is almost always 0–1 waiters, where inline-32 is waste. A clean follow-up is making WaiterList generic over its inline size and giving the consumer list a tiny one; left out here to keep the change focused.

Test plan

• cargo test -p kio — 6 passed (4 new + is_last + kio: rework Producer::poll/wait to a read-only predicate that returns a Mut #1735's poll_gates)
• clippy (nix) + fmt clean
• moq-net builds; its 143 model tests pass (exercise the unused() path)

(Written by Claude)

[coderabbitai]

No actionable comments were generated in the recent review. 🎉

ℹ️ Recent review info

⚙️ Run configuration

Configuration used: Organization UI

Review profile: CHILL

Plan: Pro

Run ID: 323e0a5d-f00b-4225-9d2e-f34c41e0f943

📥 Commits

Reviewing files that changed from the base of the PR and between 520e79c and f7d4fa8.

📒 Files selected for processing (6)

• rs/kio/CHANGELOG.md
• rs/kio/src/consumer.rs
• rs/kio/src/lib.rs
• rs/kio/src/producer.rs
• rs/kio/src/tests.rs
• rs/kio/src/weak.rs

---

Walkthrough

The change replaces a single unified WaiterList on the shared State<T> struct with three category-specific lists: waiters_value, waiters_closed, and waiters_consumer. A new take_close_waiters() method drains all three lists at once for close-event dispatch. Every poll and drop site across Producer, Consumer, and Weak is updated to register and wake only the matching category list. Mut::drop unconditionally wakes waiters_value and additionally wakes waiters_closed and waiters_consumer when a mutation closes the channel. Four regression tests using a CountWaker helper verify that events in one category do not spuriously wake waiters in another.

🚥 Pre-merge checks | ✅ 4 | ❌ 1

❌ Failed checks (1 warning)

Check name	Status	Explanation	Resolution
Linked Issues check	⚠️ Warning	The linked issue #39 concerns license change to MIT, which is completely unrelated to this PR's waiter-splitting performance optimization.	Verify the correct linked issue; this PR should reference performance or waiter-related issues, not a license change issue.

✅ Passed checks (4 passed)

Check name	Status	Explanation
Title check	✅ Passed	Title accurately summarizes the main change: splitting waiters by condition to prevent spurious wakes from writes affecting irrelevant waiter lists.
Description check	✅ Passed	Description is comprehensive and directly related to the changeset, explaining the problem, solution, design rationale, and testing approach.
Out of Scope Changes check	✅ Passed	All changes (waiter list splitting, test additions, CHANGELOG updates) are directly related to the stated objective of optimizing waiter notification behavior.
Docstring Coverage	✅ Passed	Docstring coverage is 100.00% which is sufficient. The required threshold is 80.00%.

_{✏️ Tip: You can configure your own custom pre-merge checks in the settings.}

✨ Finishing Touches

✨ Simplify code

• Create PR with simplified code
• Commit simplified code in branch claude/tender-ellis-ca7120

---

Thanks for using CodeRabbit! It's free for OSS, and your support helps us grow. If you like it, consider giving us a shout-out.

❤️ Share

• X
• Mastodon
• Reddit
• LinkedIn

_{Comment @coderabbitai help to get the list of available commands and usage tips.}